Characterizing Global Hand-Crafted Feature Descriptors for Sketch-Based Image Retrie

Abbas M.H.; Che Ani A.I.; Abd Samat A.A.; Hazim M.A.; Mohamed Mydin Hj M.Abdul Kader; Setumin S.; Muhammad Faisal Hamidi @ Abdul Rani

Characterizing Global Hand-Crafted Feature Descriptors for Sketch-Based Image Retrie

Journal

2021 6th IEEE International Conference on Recent Advances and Innovations in Engineering, ICRAIE 2021

Date Issued

2022-02-14

Author(s)

Abbas M.H.

Che Ani A.I.

Abd Samat A.A.

Hazim M.A.

Mohamed Mydin Hj M.Abdul Kader

Universiti Malaysia Perlis

Setumin S.

Muhammad Faisal Hamidi @ Abdul Rani

Universiti Malaysia Perlis

DOI

10.1109/ICRAIE52900.2021.9703823

Handle (URI)

https://hdl.handle.net/20.500.14170/8221

Abstract

Sketch-based Image Retrieval (SBIR) is an essential method of searching objects by drawing them in two-dimensional (2-D). The SBIR problem was that the image inside the dataset did not match the pair due to its modality difference. The main objective of this research is to compare which descriptor is better adapted to the SBIR. Histogram of Oriented Gradient (HOG), Gabor, and Local Binary Pattern (LBP) are selected for the study. Each sample will be divided into distance measures with features vector that corresponds to the training dataset will be compared the performance in order to study the best adapted to SBIR. Histogram of Oriented Gradient (HOG) is identified as the best descriptor for SBIR. HOG has shown a better performance than Gabor and LBP since the method shows the highest percent of distance matrices accuracy.